Visual segmentation of "simple" objects for robots

Abstract

The ability to automatically segment a "simple" object of any size from its background is important for an active agent (e.g. a robot) to interact effectively in the real world. Recently, we proposed an algorithm [14] to segment a "simple" object in a scale invariant manner given a point anywhere inside that object. However, we did not provide a strategy to select a point inside a "simple" object in [14]. In this paper, we propose a new system that automatically selects the points inside different "simple" objects in the scene, carries out the segmentation process for the selected points, and outputs only the regions corresponding to the "simple" objects in the scene. The proposed attention mechanism for the segmentation problem utilizes, for the first time, the concept of border ownership [17].